Dynamic Programming Approaches for Estimating and Applying Large-scale Discrete Choice Models

People go through their life making all kinds of decisions, and some of these decisions affect their demand for transportation, for example, their choices of where to live and where to work, how and when to travel and which route to take. Transport related choices are typically time dependent and characterized by large number of alternatives that can be spatially correlated. This thesis deals with models that can be used to analyze and predict discrete choices in large-scale networks. The proposed models and methods are highly relevant for, but not limited to, transport applications. We model decisions as sequences of choices within the dynamic discrete choice framework, also known as parametric Markov decision processes. Such models are known to be difficult to estimate and to apply to make predictions because dynamic programming problems need to be solved in order to compute choice probabilities. In this thesis we show that it is possible to explore the network structure and the flexibility of dynamic programming so that the dynamic discrete choice modeling approach is not only useful to model time dependent choices, but also makes it easier to model large-scale static choices. The thesis consists of seven articles containing a number of models and methods for estimating, applying and testing large-scale discrete choice models. In the following we group the contributions under three themes: route choice modeling, large-scale multivariate extreme value (MEV) model estimation and nonlinear optimization algorithms. Five articles are related to route choice modeling. We propose different dynamic discrete choice models that allow paths to be correlated based on the MEV and mixed logit models. The resulting route choice models become expensive to estimate and we deal with this challenge by proposing innovative methods that allow to reduce the estimation cost. For example, we propose a decomposition method that not only opens up for possibility of mixing, but also speeds up the estimation for simple logit models, which has implications also for traffic simulation. Moreover, we compare the utility maximization and regret minimization decision rules, and we propose a misspecification test for logit-based route choice models. The second theme is related to the estimation of static discrete choice models with large choice sets. We establish that a class of MEV models can be reformulated as dynamic discrete choice models on the networks of correlation structures. These dynamic models can then be estimated quickly using dynamic programming techniques and an efficient nonlinear optimization algorithm. Finally, the third theme focuses on structured quasi-Newton techniques for estimating discrete choice models by maximum likelihood. We examine and adapt switching methods that can be easily integrated into usual optimization algorithms (line search and trust region) to accelerate the estimation process. The proposed dynamic discrete choice models and estimation methods can be used in various discrete choice applications. In the area of big data analytics, models that can deal with large choice sets and sequential choices are important. Our research can therefore be of interest in various demand analysis applications (predictive analytics) or can be integrated with optimization models (prescriptive analytics). Furthermore, our studies indicate the potential of dynamic programming techniques in this context, even for static models, which opens up a variety of future research directions.

Investigation of the effects of flow conditions at rotor inlet on mixed flow turbine performance for automotive applications

Current trends in the automotive industry have placed increased importance on engine downsizing for passenger vehicles. Engine downsizing often results in reduced power output and turbochargers have been relied upon to restore the power output and maintain drivability. As improved power output is required across a wide range of engine operating conditions, it is necessary for the turbocharger to operate effectively at both design and off-design conditions. One off-design condition of considerable importance for turbocharger turbines is low velocity ratio operation, which refers to the combination of high exhaust gas velocity and low turbine rotational speed. Conventional radial flow turbines are constrained to achieve peak efficiency at the relatively high velocity ratio of 0.7, due the requirement to maintain a zero inlet blade angle for structural reasons. Several methods exist to potentially shift turbine peak efficiency to lower velocity ratios. One method is to utilize a mixed flow turbine as an alternative to a radial flow turbine. In addition to radial and circumferential components, the flow entering a mixed flow turbine also has an axial component. This allows the flow to experience a non-zero inlet blade angle, potentially shifting peak efficiency to a lower velocity ratio when compared to an equivalent radial flow turbine.
This study examined the effects of varying the flow conditions at the inlet to a mixed flow turbine and evaluated the subsequent impact on performance. The primary parameters examined were average inlet flow angle, the spanwise distribution of flow angle across the inlet and inlet flow cone angle. The results have indicated that the inlet flow angle significantly influenced the degree of reaction across the rotor and the turbine efficiency. The rotor studied was a custom in-house design based on a state-of-the-art radial flow turbine design. A numerical approach was used as the basis for this investigation and the numerical model has been validated against experimental data obtained from the cold flow turbine test rig at Queen’s University Belfast. The results of the study have provided a useful insight into how the flow conditions at rotor inlet influence the performance of a mixed flow turbine.

Dynamic Programming Approaches for Estimating and Applying Large-scale Discrete Choice Models

People go through their life making all kinds of decisions, and some of these decisions affect their demand for transportation, for example, their choices of where to live and where to work, how and when to travel and which route to take. Transport related choices are typically time dependent and characterized by large number of alternatives that can be spatially correlated. This thesis deals with models that can be used to analyze and predict discrete choices in large-scale networks. The proposed models and methods are highly relevant for, but not limited to, transport applications. We model decisions as sequences of choices within the dynamic discrete choice framework, also known as parametric Markov decision processes. Such models are known to be difficult to estimate and to apply to make predictions because dynamic programming problems need to be solved in order to compute choice probabilities. In this thesis we show that it is possible to explore the network structure and the flexibility of dynamic programming so that the dynamic discrete choice modeling approach is not only useful to model time dependent choices, but also makes it easier to model large-scale static choices. The thesis consists of seven articles containing a number of models and methods for estimating, applying and testing large-scale discrete choice models. In the following we group the contributions under three themes: route choice modeling, large-scale multivariate extreme value (MEV) model estimation and nonlinear optimization algorithms. Five articles are related to route choice modeling. We propose different dynamic discrete choice models that allow paths to be correlated based on the MEV and mixed logit models. The resulting route choice models become expensive to estimate and we deal with this challenge by proposing innovative methods that allow to reduce the estimation cost. For example, we propose a decomposition method that not only opens up for possibility of mixing, but also speeds up the estimation for simple logit models, which has implications also for traffic simulation. Moreover, we compare the utility maximization and regret minimization decision rules, and we propose a misspecification test for logit-based route choice models. The second theme is related to the estimation of static discrete choice models with large choice sets. We establish that a class of MEV models can be reformulated as dynamic discrete choice models on the networks of correlation structures. These dynamic models can then be estimated quickly using dynamic programming techniques and an efficient nonlinear optimization algorithm. Finally, the third theme focuses on structured quasi-Newton techniques for estimating discrete choice models by maximum likelihood. We examine and adapt switching methods that can be easily integrated into usual optimization algorithms (line search and trust region) to accelerate the estimation process. The proposed dynamic discrete choice models and estimation methods can be used in various discrete choice applications. In the area of big data analytics, models that can deal with large choice sets and sequential choices are important. Our research can therefore be of interest in various demand analysis applications (predictive analytics) or can be integrated with optimization models (prescriptive analytics). Furthermore, our studies indicate the potential of dynamic programming techniques in this context, even for static models, which opens up a variety of future research directions.

Mathematical Programming Models for Influence Maximization on Social Networks

In this dissertation, we apply mathematical programming techniques (i.e., integer programming and polyhedral combinatorics) to develop exact approaches for influence maximization on social networks. We study four combinatorial optimization problems that deal with maximizing influence at minimum cost over a social network. To our knowl- edge, all previous work to date involving influence maximization problems has focused on heuristics and approximation. We start with the following viral marketing problem that has attracted a significant amount of interest from the computer science literature. Given a social network, find a target set of customers to seed with a product. Then, a cascade will be caused by these initial adopters and other people start to adopt this product due to the influence they re- ceive from earlier adopters. The idea is to find the minimum cost that results in the entire network adopting the product. We first study a problem called the Weighted Target Set Selection (WTSS) Prob- lem. In the WTSS problem, the diffusion can take place over as many time periods as needed and a free product is given out to the individuals in the target set. Restricting the number of time periods that the diffusion takes place over to be one, we obtain a problem called the Positive Influence Dominating Set (PIDS) problem. Next, incorporating partial incentives, we consider a problem called the Least Cost Influence Problem (LCIP). The fourth problem studied is the One Time Period Least Cost Influence Problem (1TPLCIP) which is identical to the LCIP except that we restrict the number of time periods that the diffusion takes place over to be one. We apply a common research paradigm to each of these four problems. First, we work on special graphs: trees and cycles. Based on the insights we obtain from special graphs, we develop efficient methods for general graphs. On trees, first, we propose a polynomial time algorithm. More importantly, we present a tight and compact extended formulation. We also project the extended formulation onto the space of the natural vari- ables that gives the polytope on trees. Next, building upon the result for trees---we derive the polytope on cycles for the WTSS problem; as well as a polynomial time algorithm on cycles. This leads to our contribution on general graphs. For the WTSS problem and the LCIP, using the observation that the influence propagation network must be a directed acyclic graph (DAG), the strong formulation for trees can be embedded into a formulation on general graphs. We use this to design and implement a branch-and-cut approach for the WTSS problem and the LCIP. In our computational study, we are able to obtain high quality solutions for random graph instances with up to 10,000 nodes and 20,000 edges (40,000 arcs) within a reasonable amount of time.

Augmented Mixed Models For Clustered Proportion Data.

Often in biomedical research, we deal with continuous (clustered) proportion responses ranging between zero and one quantifying the disease status of the cluster units. Interestingly, the study population might also consist of relatively disease-free as well as highly diseased subjects, contributing to proportion values in the interval [0, 1]. Regression on a variety of parametric densities with support lying in (0, 1), such as beta regression, can assess important covariate effects. However, they are deemed inappropriate due to the presence of zeros and/or ones. To evade this, we introduce a class of general proportion density, and further augment the probabilities of zero and one to this general proportion density, controlling for the clustering. Our approach is Bayesian and presents a computationally convenient framework amenable to available freeware. Bayesian case-deletion influence diagnostics based on q-divergence measures are automatic from the Markov chain Monte Carlo output. The methodology is illustrated using both simulation studies and application to a real dataset from a clinical periodontology study.

Mixed rain forest in southeastern Brazil: tree species regeneration and floristic relationships in a remaining stretch of forest near the city of Itaberá, Brazil

The aim of this work was to evaluate the floristic composition, richness, and diversity of the upper and lower strata of a stretch of mixed rain forest near the city of Itaberá, in southeastern Brazil. We also investigated the differences between this conservation area and other stretches of mixed rain forest in southern and southeastern Brazil, as well as other nearby forest formations, in terms of their floristic relationships. For our survey of the upper stratum (diameter at breast height [DBH] > 15 cm), we established 50 permanent plots of 10 × 20 m. Within each of those plots, we designated five, randomly located, 1 × 1 m subplots, in order to survey the lower stratum (total height > 30 cm and DBH < 15 cm). In the upper stratum, we sampled 1429 trees and shrubs, belonging to 134 species, 93 genera, and 47 families. In the lower stratum, we sampled 758 trees and shrubs, belonging to 93 species, 66 genera, and 39 families. In our floristic and phytosociological surveys, we recorded 177 species, belonging to 106 genera and 52 families. The Shannon Diversity Index was 4.12 and 3.5 for the upper and lower strata, respectively. Cluster analysis indicated that nearby forest formations had the strongest floristic influence on the study area, which was therefore distinct from other mixed rain forests in southern Brazil and in the Serra da Mantiqueira mountain range.

Influence of separate and mixed experimental designs on reaction times to two simple visual stimuli

Testing contexts have been shown to critically influence experimental results in psychophysical studies. One of these contexts that show important modulation of the behavioral effects of different stimulatory conditions is the separate (blocked) or mixed presentation of these stimulatory conditions. The study presents evidence that the apparent discriminabilities of two target stimuli can change according to which of these two testing contexts is used. A cross inside a ring and a vertical line inside a ring were presented as go stimuli in a go/no-go reaction time task. In one experiment, each of these stimuli was presented to a different group of volunteers and in another experiment they were presented to the same group of volunteers, randomly mixed in the blocks of trials. Similar reaction times were obtained for the two stimuli in the first experiment, and different reaction times (faster for the cross) in the second experiment. The latter result indicates that the two stimuli have different discriminabilities from the no-go stimulus; the cross having greater discriminability. This difference is however masked, presumably by the adoption of specific compensatory attentional sets, in a separate testing context.

The effect of using mixed initiator systems on the efficiency of photopolymerization of dental resins

A study was performed in order to determine the efficiency of the simultaneous use of the photoinitiators phenylpropanedione (PPD) and camphorquinone (CQ) in the polymerization of acrylic polymers and evaluate possible mechanisms leading to synergism or antagonism. It was found that efficiencies of both initiators taken individually are higher than that of their mixture, indicating that when both dyes are used simultaneously there will be an energy transfer from the more efficient initiator (CQ) to the less efficient one (PPD). Also, there was no proof of any reaction between the amine present in the CQ formulation and the PPD excited state.

A mixed colony of Scaptotrigona depilis and Nannotrigona testaceicornis (Hymenoptera, Apidae, Meliponina)

We describe a case of a spontaneously established mixed colony of two species of stingless bees. The host colony of Scaptotrigona depilis, an aggressive bee that forms large colonies, was invaded by workers of Nannotrigona testaceicornis, a smaller bee that forms small colonies. The host colony and the invading species colony were maintained in next boxes about 1.5 m apart. The N. testaceicornis colony had been recently divided. Observations were made daily for 10 min, and every two weeks the colony was opened for observations within the nest. Initially the host colony bees repulsed the invading species, but as their numbers built up, they were no longer able to defend the entrance. An estimated 60-90 N. testaceicornis workers lived integrated into the colony of S. depilis for 58 days. During this period, they reconstructed and maintained the entrance tube, changing it to an entrance typical of N. testaceicornis. They also collected food and building material for the host colony. Nannotrigona testaceicornis tolerated transit of S. depilis through the entrance, but did not allow the host species to remain within the tube, though the attacks never resulted in bee mortality. Aggression was limited to biting the wings; when the bees fell to the ground they immediately separated and flew back. There have been very few reports of spontaneously occurring mixed stingless bee colonies. It is difficult to determine what caused the association that we found; probably workers of N. testaceicornis got lost when we split their colony, and then they invaded the colony of S. depilis.

Phylogenetic position of the acariform mites: sensitivity to homology assessment under total evidence

Background: Mites (Acari) have traditionally been treated as monophyletic, albeit composed of two major lineages: Acariformes and Parasitiformes. Yet recent studies based on morphology, molecular data, or combinations thereof, have increasingly drawn their monophyly into question. Furthermore, the usually basal (molecular) position of one or both mite lineages among the chelicerates is in conflict to their morphology, and to the widely accepted view that mites are close relatives of Ricinulei. Results: The phylogenetic position of the acariform mites is examined through employing SSU, partial LSU sequences, and morphology from 91 chelicerate extant terminals (forty Acariformes). In a static homology framework, molecular sequences were aligned using their secondary structure as guide, whereby regions of ambiguous alignment were discarded, and pre-aligned sequences analyzed under parsimony and different mixed models in a Bayesian inference. Parsimony and Bayesian analyses led to trees largely congruent concerning infraordinal, well-supported branches, but with low support for inter-ordinal relationships. An exception is Solifugae + Acariformes (P. P = 100%, J. = 0.91). In a dynamic homology framework, two analyses were run: a standard POY analysis and an analysis constrained by secondary structure. Both analyses led to largely congruent trees; supporting a (Palpigradi (Solifugae Acariformes)) clade and Ricinulei as sister group of Tetrapulmonata with the topology (Ricinulei (Amblypygi (Uropygi Araneae))). Combined analysis with two different morphological data matrices were run in order to evaluate the impact of constraining the analysis on the recovered topology when employing secondary structure as a guide for homology establishment. The constrained combined analysis yielded two topologies similar to the exclusively molecular analysis for both morphological matrices, except for the recovery of Pedipalpi instead of the (Uropygi Araneae) clade. The standard (direct optimization) POY analysis, however, led to the recovery of trees differing in the absence of the otherwise well-supported group Solifugae + Acariformes. Conclusions: Previous studies combining ribosomal sequences and morphology often recovered topologies similar to purely morphological analyses of Chelicerata. The apparent stability of certain clades not recovered here, like Haplocnemata and Acari, is regarded as a byproduct of the way the molecular homology was previously established using the instrumentalist approach implemented in POY. Constraining the analysis by a priori homology assessment is defended here as a way of maintaining the severity of the test when adding new data to the analysis. Although the strength of the method advocated here is keeping phylogenetic information from regions usually discarded in an exclusively static homology framework; it still has the inconvenience of being uninformative on the effect of alignment ambiguity on resampling methods of clade support estimation. Finally, putative morphological apomorphies of Solifugae + Acariformes are the reduction of the proximal cheliceral podomere, medial abutting of the leg coxae, loss of sperm nuclear membrane, and presence of differentiated germinative and secretory regions in the testis delivering their products into a common lumen.

Radiative decay of the X(3872) as a mixed molecule-charmonium state in QCD sum rules

We use QCD sum rules (QCDSR) to calculate the width of the radiative decay of the meson X(3872), assumed to be a mixture between charmonium and exotic molecular [c (q) over bar][q (c) over bar] states with J(PC) = 1(++). We find that in a small range for the values of the mixing angle, 5 degrees <= theta <= 13 degrees, we get the branching ratio Gamma(X -> J/psi gamma)/Gamma(X -> J/psi pi(+)pi(-)) = 0.19 +/- 0.13, which is in agreement, with the experimental value. This result is compatible with the analysis of the mass and decay width of the mode J/psi(n pi) performed in the same approach.

QCD sum rules for the X(3872) as a mixed molecule-charmonium state

We use QCD sum rules to test the nature of the meson X(3872), assumed to be a mixture between charmonium and exotic molecular [c (q) over bar][q (c) over bar] states with J(PC) = 1(++). We find that there is only a small range for the values of the mixing angle theta that can provide simultaneously good agreement with the experimental value of the mass and the decay width, and this range is 5(0) <= theta <= 3(0). In this range we get m(X) = (3.77 +/- 0.18) GeV and Gamma(X -> J/psi pi(+)pi(-)) = (9.3 +/- 6.9) MeV, which are compatible, within the errors, with the experimental values. We, therefore, conclude that the X(3872) is approximately 97% a charmonium state with 3% admixture of similar to 88% D(0)D*(0) molecule and similar to 12% D(+)D*(-) molecule.

Cold nuclear matter effects on J/psi production as constrained by deuteron-gold measurements at root S(NN)=200 GeV

We present a new analysis of J/psi production yields in deuteron-gold collisions at root s(NN) =200 GeV using data taken from the PHENIX experiment in 2003 and previously published in S. S. Adler [Phys. Rev. Lett 96, 012304 (2006)]. The high statistics proton-proton J/psi data taken in 2005 are used to improve the baseline measurement and thus construct updated cold nuclear matter modification factors (R(dAu)). A suppression of J/psi in cold nuclear matter is observed as one goes forward in rapidity (in the deuteron-going direction), corresponding to a region more sensitive to initial-state low-x gluons in the gold nucleus. The measured nuclear modification factors are compared to theoretical calculations of nuclear shadowing to which a J/psi (or precursor) breakup cross section is added. Breakup cross sections of sigma(breakup)=2.8(-1.4)(+1.7) (2.2(-1.5)(+1.6)) mb are obtained by fitting these calculations to the data using two different models of nuclear shadowing. These breakup cross-section values are consistent within large uncertainties with the 4.2 +/- 0.5 mb determined at lower collision energies. Projecting this range of cold nuclear matter effects to copper-copper and gold-gold collisions reveals that the current constraints are not sufficient to firmly quantify the additional hot nuclear matter effect.

Detection of catechol using mixed Langmuir-Blodgett films of a phospholipid and phthalocyanines as voltammetric sensors

The combination of metallic phthalocyanines (MPcs) and biomolecules has been explored in the literature either as mimetic systems to investigate molecular interactions or as supporting layers to immobilize biomolecules. Here, Langmuir-Blodgett (LB) films containing the phospholipid dimyristoyl phosphatidic acid (DMPA) mixed either with iron phthalocyanine (FePc) or with lutetium bisphthalocyanine (LuPc(2)) were applied as ITO modified-electrodes in the detection of catechol using cyclic voltammetry. The mixed Langmuir films of FePc + DMPA and LuPc(2) + DMPA displayed surface-pressure isotherms with no evidence of molecular-level interactions. The Fourier Transform Infrared (FTIR) spectra of the multilayer LB films confirmed the lack of interaction between the components. The DMPA and the FePc molecules were found to be oriented perpendicularly to the substrate, while LuPc(2) molecules were randomly organized. The phospholipid matrix induced a remarkable electrocatalytic effect on the phthalocyanines; as a result the mixed LB films deposited on ITO could be used to detect catechol with detection limits of 4.30 x 10(-7) and 3.34 x 10(-7) M for FePc + DMPA and LuPc(2) + DMPA, respectively. Results from kinetics experiments revealed that ion diffusion dominated the response of the modified electrodes. The sensitivity was comparable to that of other non-enzymatic sensors, which is sufficient to detect catechol in the food industry. The higher stability of the electrochemical response of the LB films and the ability to control the molecular architecture are promising for further studies with incorporation of biomolecules.